Disobedience rate: 11%, original: 74%
KL divergence: 0.0126
Parameters:
direction_index = per layer
attn.o_proj.max_weight = 1.40
attn.o_proj.max_weight_position = 16.90
attn.o_proj.min_weight = 0.23
attn.o_proj.min_weight_distance = 7.93
mlp.down_proj.max_weight = 1.21
mlp.down_proj.max_weight_position = 23.86
mlp.down_proj.min_weight = 0.11
mlp.down_proj.min_weight_distance = 5.50
- Downloads last month
- 2